A key object discovery method based on spatial trajectory big data analysis

Abstract

This invention relates to a method for identifying key targets based on spatial trajectory big data analysis, belonging to the field of data processing technology. The method includes: responding to acquired anonymized object data; comparing the anonymized object data with a behavioral habit model; and obtaining suspected key targets based on the comparison results; generating movement trajectory models of the suspected key targets; comparing the similarity of movement trajectory models of any two suspected key targets; and determining key targets and generating early warnings based on the similarity comparison results; the similarity comparison includes the overlap degree of movement trajectory models, the amount of overlap in movement trajectory models, and the temporal correlation between the overlapping areas of movement trajectory models and hot events. This invention discovers potential risk factors by analyzing and comparing the acquired data. This method has better proactiveness and also solves the problem of increased management costs caused by ineffective early warnings.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a key object discovery method based on spatial trajectory big data analysis. Background Art

[0002] Data mining refers to the process of searching for information hidden in a large amount of data through algorithms. Mining based on historical data can obtain rich information. Taking public safety and social security as an example, through the mining of big data, potential risk factors can be discovered and eliminated to a certain extent, which helps to reduce the overall management cost of society and the scope of negative events.

[0003] As for how to discover potential risk factors, the commonly used technical means include establishing a database and early warning mechanism. The database stores potential dangerous objects, and when the behavior of these dangerous objects triggers an early warning, they will be checked in a timely manner.

[0004] However, it should be clear that although the database stores massive amounts of data, it lacks effective big data mining algorithms and practical models, making it impossible to discover potential risk factors in advance, nor to dynamically associate them with warnings to reduce invalid warnings. Summary of the invention

[0005] The present invention provides a key object discovery method based on spatial trajectory big data analysis, which discovers potential risk factors by performing spatiotemporal analysis and comparison on the acquired data. This method has better preemptiveness and also solves the problem of increased management costs caused by invalid early warnings.

[0006] The above object of the present invention is achieved through the following technical solutions:

[0007] A key object discovery method based on spatial trajectory big data analysis, comprising:

[0008] In response to the acquired object anesthetized data, the object anesthetized data is compared with the behavior habit model, and the suspected key object is obtained according to the comparison result;

[0009] Generate a movement trajectory model of suspected key objects;

[0010] Comparing the similarity of the movement trajectory models of any two suspected key objects; and

[0011] Obtain key objects based on similarity comparison results and generate warnings;

[0012] The similarity comparison includes the overlap degree of the movement trajectory models, the overlap amount of the movement trajectory models, and the temporal correlation between the overlap area of ​​the movement trajectory models and the hot event.

[0013] In one possible implementation, comparing the object desensitized data with the behavior habit model includes:

[0014] assigning an identity type to the object-masked data according to the object-masked data; and

[0015] The object's desensitized data is compared with the corresponding identity type sub-model.

[0016] In one possible implementation, comparing the object desensitized data with the behavior habit model includes:

[0017] Classify an object's desensitized data according to the dimension;

[0018] Construct multiple judgment sub-models, each of which includes multiple dimensions, and each judgment sub-model has at least two dimensions with different weights;

[0019] The judgment sub-model evaluates the object's desensitized data; and

[0020] Assigning an identity type to the object's desensitized data according to the evaluation result and comparing the object's desensitized data with the corresponding identity type sub-model;

[0021] In each evaluation process, the dimensions in the judgment sub-model belong to the same identity type.

[0022] In one possible implementation, the criteria for assigning identity types include:

[0023] For the same identity type, an assessment is performed, and the proportion of the assigned results given by the judgment sub-model in the results given by all judgment sub-models is greater than or equal to the set proportion;

[0024] Alternatively, multiple assessments are performed on the same identity type, and the percentage of the assigned times in the total number of times is greater than or equal to the set ratio.

[0025] In a possible implementation, comparing the similarity of the movement trajectory models of any two suspected key objects includes:

[0026] The movement trajectory model of the suspected key object is displayed on the map by connecting the location points, the location points include the actual location and the virtual location, and the actual location and the virtual location are connected in time sequence;

[0027] Compare the similarity of the lines connecting the location points of two suspected key objects.

[0028] In a possible implementation, the location point includes an actual range and a virtual range;

[0029] The similarities include:

[0030] The actual ranges and / or virtual ranges of the locations of two suspected key objects overlap;

[0031] There is an intersection between the lines connecting the two suspected key objects;

[0032] The actual range and / or virtual range of a location point of a suspected key object overlaps with a line connecting another suspected key object.

[0033] In a possible implementation, when there is a missing movement trajectory model of a suspected key object, the following is further included:

[0034] The potential range is constructed based on the two endpoints and the missing time of the missing part of the movement trajectory model;

[0035] Select multiple suspected locations within the potential range; and

[0036] Fill in the missing parts of the movement trajectory model according to the selected suspected position points;

[0037] Among them, the number of times of selecting multiple suspected location points is multiple times;

[0038] Fill in the missing part of each pair of movement trajectory models, and compare the similarity between the movement trajectory model and the movement trajectory model of any suspected key object;

[0039] During multiple comparisons, the determined position points are selected from the suspected position points and the determined position points are used to fill in the missing parts of the movement trajectory model.

[0040] The beneficial effects of the present invention are:

[0041] 1. The present invention uses a similarity comparison judgment method, that is, generating an early warning by judging whether multiple suspected key objects are clustered. The advantage of this method is that it improves the accuracy of the judgment, because compared with the individual actions of a suspected key object, the group activities of multiple suspected key objects are obviously more dangerous. When the group activities of multiple suspected key objects are found, an early warning can be directly generated.

[0042] 2. The similarity comparison judgment method of the present invention is also proactive, that is, when group activities are discovered and the behavioral habit model does not respond, it can also generate an early warning more accurately. At the same time, it can also reduce the false alarm rate of verification and achieve effective control of the overall social management cost.

[0043] In summary, the present invention discovers potential risk factors by analyzing and comparing the acquired data. This method has better proactiveness and also solves the problem of increased management costs caused by invalid early warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic flowchart of the steps of a key object discovery method.

[0045] Figure 2 It is a schematic diagram of using a behavioral habit model to classify object desensitized data.

[0046] Figure 3 It is a schematic diagram of the principle of comparing the movement trajectory models of two suspected key objects.

[0047] Figure 4 It is another schematic diagram of the principle of comparing the movement trajectory models of two suspected key objects.

[0048] Figure 5 It is another schematic diagram of the principle of comparing the movement trajectory models of two suspected key objects.

[0049] Figure 6 It is a schematic diagram of the process of steps to assign an identity type.

[0050] Figure 7 It is a schematic diagram of the composition of location points.

[0051] Figure 8 It is a schematic diagram of the principle of obtaining multiple suspected location points through a suspected range.

[0052] Fig. 9 is based on Figure 8 A schematic diagram of the principle of obtaining the determined position point is given.

[0053] Fig.10 is based on Fig. 9 A schematic diagram of the principle of using determined position points to fill in the missing parts of the movement trajectory model is given. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0055] See also Figure 1 , which is a key object discovery method based on spatial trajectory big data analysis, and the method includes the following steps:

[0056] S101, in response to the acquired object anesthetized data, comparing the object anesthetized data with the behavior habit model, and obtaining a suspected key object according to the comparison result;

[0057] S102, generating a movement trajectory model of a suspected key object;

[0058] S103, comparing the similarity of the movement trajectory models of any two suspected key objects, and determining the key object according to the similarity comparison result; and

[0059] S104, generating an early warning for key targets;

[0060] The similarity comparison includes the overlap degree of the movement trajectory models, the overlap amount of the movement trajectory models, and the temporal correlation between the overlap area of ​​the movement trajectory models and the hot event.

[0061] Specifically, in step S101, object desensitized data will be received. These object desensitized data are generated based on multiple objects, that is, each object will generate a certain amount of data in social activities (work, social, transportation, communication, etc.), and these data can reflect some behavioral characteristics and intentions of the object to a certain extent.

[0062] See also Figure 2 After receiving the object desensitized data, the object desensitized data is compared with the behavioral habit model, and the normal object and the suspected key object are obtained according to the comparison results. The behavioral habit model refers to a model built according to the characteristics of different objects (age, occupation, hobbies, habits, etc.). For a fixed occupation, a fixed group or a group of people in a certain age group, they will have some similar characteristics. By identifying these characteristics, it is possible to reversely deduce whether the object needs to be included in the suspected key object.

[0063] It should be noted here that for object-massive data, after data collection, data cleaning and data desensitization, the object-massive data obtained is not targeted and cannot be directly used for personnel identification. Therefore, analysis and judgment are required to obtain the required results. Therefore, it is necessary to use the participation of behavioral habit models to conduct preliminary screening of object-massive data.

[0064] Then, step S102 is performed, in which a movement trajectory model of the suspected key object is generated, and the movement trajectory model represents the movement of the suspected key object within the spatial range. In some examples, the generation of the movement trajectory model of the suspected key object is to represent the movement of the suspected key object within the spatial range in the form of a trajectory, such as the trajectory generated when walking from point A to point B.

[0065] Then, step S103 is executed, in which the similarity of the movement trajectory models of any two suspected key objects is compared, and the key object is determined according to the similarity comparison result. The similarity comparison results are similar and dissimilar. If the comparison result is similar, it is determined as a key object.

[0066] The purpose of the comparison here is to find out whether there is overlap in the movement of suspected key objects in the spatial range and whether there is overlap with a hot event in the time dimension. Hot events are mainly public safety events, and the hot events are determined and input by the operator here.

[0067] Comparing the similarity of the movement trajectory models of any two suspected key objects has the following dimensions:

[0068] (1) The overlap of the trajectory models. The overlap of the trajectory models refers to the overlap area or overlap length of two trajectory models in the spatial range. Figure 3 As shown;

[0069] (2) The overlap of the trajectory models refers to the number of overlapping points of two trajectory models in the spatial range. Figure 4 As shown;

[0070] (3) The temporal correlation between the overlapping area of ​​the movement trajectory model and the hotspot event refers to the overlap between two suspected key objects in the spatial range and a hotspot event in the temporal dimension. Figure 5 shown.

[0071] Finally, step S104 is executed, in which a key object warning is generated for the selected key objects.

[0072] In step S101, the comparison between the object desensitized data and the behavior habit model has three results, namely, the desensitized data belonging to an object matches the behavior habit model of a normal object, the desensitized data belonging to an object matches the behavior habit model of a suspected key object, and the desensitized data belonging to an object is not in the known behavior habit model (cannot match), Figure 2 shown.

[0073] For the second and third results, a suspected key target warning can be directly generated. However, as mentioned in the previous article, the problem with this type of warning is that if personnel are not dispatched for verification, the generated warning is worthless; and too many invalid warnings will increase additional overall social management costs.

[0074] In the present invention, a similarity comparison judgment method is used, that is, an early warning is generated by judging whether multiple suspected key objects are clustered. The advantage of this method is that it improves the accuracy of the judgment, because compared with the individual actions of a suspected key object, the group activities of multiple suspected key objects are obviously more dangerous. When the group activities of multiple suspected key objects are found, an early warning can be directly generated.

[0075] Moreover, this judgment method is also proactive, that is, when group activities are discovered and the behavioral habit model does not respond, it can generate an early warning more accurately. At the same time, it can also reduce the false alarm rate of verification and achieve effective control of the overall social management costs.

[0076] Some examples of comparing a subject's desensitized data to a behavioral habit model include:

[0077] S201, assigning an identity type to the object desensitized data according to the object desensitized data; and

[0078] S202, comparing the object desensitized data with the corresponding identity type sub-model.

[0079] Specifically, in step S201 and step S202, the object desensitized data is classified according to the identity type, and then the object desensitized data is compared with the corresponding identity type sub-model. The advantage of this method is that the object desensitized data can be preliminarily divided by identity type, and the classification of identity type is more targeted.

[0080] In some possible implementations, the identity type may be an occupation type (eg, teacher), or a definition of a specific group of people, such as office workers.

[0081] For the identity type sub-model, the sub-model is defined in advance and is directly called during the comparison process.

[0082] See also Figure 6 When ambiguity occurs in assigning identity types and the identity type cannot be assigned directly, use the following method to handle it:

[0083] S301, classifying an object's desensitized data according to the dimension;

[0084] S302, constructing multiple judgment sub-models, each judgment sub-model includes multiple dimensions, and each judgment model has at least two dimensions with different weights;

[0085] S303, the judgment sub-model evaluates the desensitized data of the object; and

[0086] S304, assigning an identity type to the object desensitized data according to the evaluation result and comparing the object desensitized data with the corresponding identity type sub-model;

[0087] In each evaluation process, the dimensions in the judgment model belong to the same identity type.

[0088] Specifically, corresponding indicator parameters will be set for each identity type. When all or part of the indicator parameters meet the requirements, the identity type of the object's desensitized data can be determined.

[0089] It should be understood that the identity type has multiple indicator parameters. When determining the identity type of the object's desensitized data, all indicator parameters may meet the requirements (for example, the number of indicator parameters is 10, and the number of indicator parameters that meet the requirements is also 10); or some indicator parameters may meet the requirements (for example, the number of indicator parameters is 10, and the number of indicator parameters that meet the requirements is also 7, 8 or 9).

[0090] The indicator parameters of identity types can be understood in this way: when a person has a certain identity type, there will be some characteristic data in the data generated by him. Taking office workers as an example, their travel methods, travel times and travel routes all have certain regularities, which can be defined as indicator parameters.

[0091] For an identity type, taking travel mode as an example, if the travel mode is consistent with the travel mode of office workers, then the indicator parameter is classified as an indicator parameter that meets the requirements.

[0092] Therefore, in step S301, the desensitized data of an object will be classified according to the dimensions, and then multiple judgment sub-models will be constructed. Each judgment sub-model includes multiple dimensions, and each judgment model has at least two dimensions with different weights, which is the content in step S302.

[0093] For dimension, it is a variable set in advance, such as commuting time, personal preferences, shopping habits, address, travel mode, etc.

[0094] The method of constructing multiple judgment sub-models is described here using random forest. Random forest adopts the idea of ​​bagging. The so-called bagging is:

[0095] (1) Each time, n training samples are taken from the training set with replacement to form a new training set;

[0096] (2) Using the new training set, train M sub-models;

[0097] (3) For classification problems, a voting method is used, and the classification category of the sub-model with the most votes is the final category; for regression problems, a simple averaging method is used to obtain the predicted value.

[0098] The judgment sub-model is constructed using random forest.

[0099] For a judgment sub-model, a dimension less than the total number of dimensions will be assigned to the judgment model, and then a weight will be assigned to each dimension. For example, for a judgment sub-model, the number of dimensions assigned is two, and the weights of the two dimensions may be 40% and 60% or 50% and 50%.

[0100] In an evaluation process, the dimensions in all judgment sub-models belong to the same identity type. Multiple judgment sub-models evaluate the object's desensitized data, and then assign an identity type to the object's desensitized data based on the evaluation results.

[0101] The evaluation method in random forest has only two results: yes or no.

[0102] The advantage of this evaluation method is that it improves the accuracy of assigning identity types to object desensitized data, because it follows the principle of minority obeys majority and reduces variance by integrating multiple judgment sub-models, thereby improving the robustness and generalization ability of the judgment sub-model.

[0103] In some possible implementations, the number of dimensions in all judgment sub-models tends to be consistent.

[0104] In some possible implementations, the weights of the number of dimensions in all judgment sub-models tend to be consistent or the weight values ​​of the same dimensions in each judgment sub-model are fixed.

[0105] There are two ways to assign identity types:

[0106] The first one is to evaluate the same identity type once, and the proportion of the assigned results given by the judgment sub-model in the results given by all judgment sub-models is greater than or equal to the set proportion;

[0107] The second is to conduct multiple evaluations on the same identity type, and the number of times assigned accounts for a proportion of the total number of times that is greater than or equal to the set proportion.

[0108] Setting the ratio is a pre-set parameter.

[0109] In some examples, comparing the similarity of the movement trajectory models of any two suspected key objects includes the following steps:

[0110] S401, displaying the movement trajectory model of the suspected key object on the map by connecting position points, where the position points include actual locations and virtual locations, and the actual locations and virtual locations are connected in chronological order;

[0111] S402, comparing the similarity of the lines connecting the position points of two suspected key objects.

[0112] Specifically, in step S401 to step S402, the movement trajectory model of the suspected key object needs to be displayed on the map using position point connection lines, that is, the movement trajectory model of the suspected key object needs to be displayed in a concrete manner.

[0113] It should be noted that, in the present invention, the location points used on the map include actual locations and virtual locations. Figure 7 The actual location shown indicates the actual position where the suspected key object has appeared in space, and the virtual location indicates the position to which the suspected key object is extended in space through some means (such as telephone and transfer).

[0114] The actual location and the virtual location are represented by points on the map, and then these points are connected in the order of their appearance to visualize the movement trajectory model of the suspected key object. Then the similarity of the lines connecting the location points of two suspected key objects is compared.

[0115] There are three ways to compare the similarity of location point connections:

[0116] The actual ranges and / or virtual ranges of the locations of two suspected key objects overlap;

[0117] There is an intersection between the lines connecting the two suspected key objects;

[0118] The actual range and / or virtual range of a location point of a suspected key object overlaps with a line connecting another suspected key object.

[0119] Use any one or more of these three methods to compare the similarity of the lines connecting the location points of two suspected key objects.

[0120] In some possible implementations, the location point includes an actual range and a virtual range. The actual range refers to the actual location where the suspected key object has appeared in space. For example, if the location is a building, then the building is the actual range of the location point; for another example, if the location is a street, then within a set length range, this length of the street is the actual range of the location point.

[0121] For the virtual range, it can be understood that the movement of the suspected key object within an actual range cannot be fully and accurately recorded, so it is necessary to use the virtual range to further determine whether the location point has moved within the actual range.

[0122] The virtual range can be seen as the overflow of the real range in the spatial dimension.

[0123] In some possible implementations, the overflow of the virtual range is positively correlated with time. For example, with the actual range as the center, the radius of the virtual range is positively correlated with the residence time of the suspected key object in the actual range.

[0124] Of course, there may be a problem that the movement trajectory model of the generated suspected key object is missing. The main reasons for the missing are data loss, data cannot be collected, and data is cleaned during the data cleaning process. For this situation, use the following method to handle it:

[0125] S501, constructing a potential range according to two endpoints and missing time of the missing part of the movement trajectory model;

[0126] S502, selecting multiple suspected location points within the potential range; and

[0127] S503, filling in the missing part of the movement trajectory model according to the selected suspected position points;

[0128] See also Figure 8 In step S501 to step S503, the two endpoints and the missing time of the missing part are first determined, and then a potential range is constructed, which represents the possible activity area of ​​the suspected key object. Then, multiple suspected location points are selected (for example, randomly selected) within the potential range, and then the missing part of the movement trajectory model is filled with the selected suspected location points.

[0129] When filling, the missing part of each pair of motion trajectory models is filled once, and the similarity between the motion trajectory model and the motion trajectory model of any suspected key object is compared. If the selected suspected position point appears in the motion trajectory model of other suspected key objects, then this suspected position point is used to fill the missing part of the motion trajectory model.

[0130] In this process, the number of suspected location points selected is multiple times. In the multiple comparison processes, a certain number of suspected location points will be obtained. After obtaining these suspected location points, the confirmed location points are screened out from the suspected location points ( Fig. 9 As shown) and use the determined position points to fill in the missing parts of the movement trajectory model ( Fig.10 shown).

[0131] During the screening process, the suspected location points can be arranged according to the number of times they appear (for example, from large to small), and several suspected location points can be selected (the first few can be selected according to the order after arrangement) as the determined location points, or suspected location points whose number of appearances exceeds the set number of times can be used as the determined location points. The time of the determined location point is referenced to its appearance time in the movement trajectory model of other suspected key objects, and one is selected from multiple times or the average is calculated.

Claims

1. A key object discovery method based on spatial trajectory big data analysis, characterized in that: include: In response to the acquired object desensitized data, the object desensitized data is compared with the behavior habit model, and a suspected key object is obtained according to the comparison result; the behavior habit model is a model constructed according to the age, occupation, hobby, and habit characteristics of different objects, and has similar characteristics for a fixed occupation, a fixed group or a group of people in a certain age group; the object desensitized data is compared with the behavior habit model, and whether the object is a suspected key object is deduced according to the similarity between the object characteristics in the object desensitized data and the object characteristics in the behavior habit model; Generate a movement trajectory model of suspected key objects; Compare the similarity of the movement trajectory models of any two suspected key objects, and determine the key object based on the similarity comparison result; as well as Generate early warnings for key targets; Among them, the similarity comparison includes the overlap degree of mobile trajectory models, the overlap amount of mobile trajectory models, and the temporal correlation between the overlap area of ​​mobile trajectory models and hot events; the overlap degree of mobile trajectory models refers to the overlap area or overlap length of two mobile trajectory models in the spatial range; the overlap amount of mobile trajectory models refers to the number of overlap points of two mobile trajectory models in the spatial range; the temporal correlation between the overlap area of ​​mobile trajectory models and hot events refers to whether two suspected key objects overlap in the spatial range and a hot event in the temporal dimension.

2. The key object discovery method based on spatial trajectory big data analysis according to claim 1 is characterized in that: Comparing the object's desensitized data with the behavioral habit model includes: assigning an identity type to the object-masked data according to the object-masked data; and The object's desensitized data is compared with the corresponding identity type sub-model.

3. The key object discovery method based on spatial trajectory big data analysis according to claim 1 is characterized in that: Comparing the object's desensitized data with the behavioral habit model includes: Classify an object's desensitized data according to the dimension; Construct multiple judgment sub-models, each of which includes multiple dimensions, and each judgment sub-model has at least two dimensions with different weights; The judgment sub-model evaluates the object's desensitized data; and Assigning an identity type to the object's desensitized data according to the evaluation result and comparing the object's desensitized data with the corresponding identity type sub-model; In each evaluation process, the dimensions in the judgment sub-model belong to the same identity type.

4. The key object discovery method based on spatial trajectory big data analysis according to claim 3 is characterized in that: The criteria for assigning identity types include: For the same identity type, an assessment is performed once, and the proportion of the assigned results given by the judgment sub-model in the results given by all judgment sub-models is greater than or equal to the set proportion; or Alternatively, multiple evaluations are performed on the same identity type, and the result is that the number of times assigned is greater than or equal to the set number of times.

5. The key object discovery method based on spatial trajectory big data analysis according to any one of claims 1 to 4, characterized in that: Comparing the similarity of the movement trajectory models of any two suspected key objects includes: The movement trajectory model of the suspected key object is displayed on the map by connecting the location points, the location points include the actual location and the virtual location, and the actual location and the virtual location are connected in time sequence; Compare the similarity of the lines connecting the location points of two suspected key objects.

6. The key object discovery method based on spatial trajectory big data analysis according to claim 5 is characterized in that: The location point includes the actual range and the virtual range; Compare the similarity of the lines connecting the location points of two suspected key objects, including: The actual ranges and / or virtual ranges of the locations of two suspected key objects overlap; There is an intersection between the lines connecting the two suspected key objects; The actual range and / or virtual range of a location point of a suspected key object overlaps with a line connecting another suspected key object.

7. The key object discovery method based on spatial trajectory big data analysis according to claim 1 is characterized in that: When the movement trajectory model of the suspected key object is missing, it also includes: The potential range is constructed based on the two endpoints and the missing time of the missing part of the movement trajectory model; Select multiple suspected locations within the potential range; and Fill in the missing parts of the movement trajectory model according to the selected suspected position points; Among them, the number of times of selecting multiple suspected location points is multiple times; Fill in the missing part of each pair of movement trajectory models, and compare the similarity between the movement trajectory model and the movement trajectory model of any suspected key object; During multiple comparisons, the determined position points are selected from the suspected position points and the determined position points are used to fill in the missing parts of the movement trajectory model.

Images